That’s Sorted

September 12, 2019

I was intrigued when I saw some coverage of a recent article in Nature (paywalled, but there’s a read-only version) that coined an intriguing phrase: information gerrymandering. The six authors come from disparate fields that are not ones I particularly follow on a regular basis – biology, economics, and environmental change. But their experiment adds an intriguing wrinkle to the more information-science and sociology-of-information circles where I tend to eavesdrop.

The authors constructed an experiment in which people were recruited to play a voting game, which the researchers ran 120 times involving 2,520 participants, yielding data to analyze mathematically. The participants were assigned parties that didn’t have platforms, just colors (purple and yellow) with an equal number of members. The goal of the game was to win the most votes. To win you had to convince people from the other side to join your party; if the was a deadlock, both parties would lose. As the authors put it, “a party is most effective when it influences the largest possible number of people just enough to flip their votes, without wasting influence on those who are already convinced.”

Here’s where things get interesting. If a willingness to compromise is unevenly distributed, those who have a lot of zealots who refuse to compromise have an edge. A pretty significant edge. However, if both sides use that strategy and avoid compromise, both sides lose.

The authors tested this by adding influencers, “zealous bots,” to the game, voices that argued against compromise. Those voices had a significant effect. If you treat the kinds of social groupings we form online and in our politics as a kind of geography of like-minded bubbles, it’s quite possible to seek the edges of those bubbles to carve out gerrymandered information groups with the help of zealotry (whether provided by bots or by actual people who strongly support one side and argue against having any truck with the other).

You might be thinking “yeah, that’s exactly what people in that other party do.” But remember, all that was at stake here wasn’t reproductive rights or government regulation or the other things we tend to get excited about. It was whether purple or yellow was better. To see if this analysis played out when the stakes were more defined, the authors ran a similar analysis on European and U.S. legislative bodies and saw a similar pattern.

The authors conclude:

our study on the voter game highlights how sensitive collective decisions are to information gerrymandering on an influence network, how easily gerrymandering can arise in realistic networks and how widespread it is in real-world networks of political discourse and legislative process. Our analysis provides a new perspective and a quantitative measure to study public discourse and collective decisions across diverse contexts.

Symmetric influence assortment allows for democratic outcomes, in which the expected vote share of a party is equal to its representation among voters; and low influence assortment allows decisions to be reached with broad consensus despite different partisan goals. A party that increases its own influence assortment relative to that of the other party by coordination, strategic use of bots or encouraging a zero-sum worldview benefits from information gerrymandering and wins a disproportionate share of the vote—that is, an undemocratic outcome. However, other parties are then incentivized to increase their own influence assortment, which leaves everyone trapped in deadlock.

Thinking about this in terms of information literacy, it seems important to bear in mind that the systems we so often use for keeping up with current events are optimized both for persuasion and for attention. Zealotry is both a way to maximize attention and to push a lot of persuasion buttons. We know from recent elections that the technology developed by Facebook and Google helps advertisers target audiences with a very fine grain. In my personal experience, it’s really hard to use social media to persuade anyone who isn’t already on board, and it’s terribly common to not just talk past each other, but to do so in insulting, alienating ways. I’m not sure what the solution is. Even Google and Facebook are struggling to find better ways to detoxify political communication. But as we help students think about evaluating information, it may be interesting to zoom out from analyzing specific messages to seeing how messaging can be influenced at scale.